Skip to content

Every Day Is Zero Day | Wiz CTO Ami Luttwak

a16z Podcast

Hosted by Erik Torenberg

46m
Transcript

Original podcast

Where to listen

Summary

Written by a model from the transcript; not yet reviewed.

This episode explores the dramatic shift in cybersecurity driven by advanced AI models, which have collapsed the time needed to write and execute complex exploits from months to mere minutes. With October dubbed Black October due to an unprecedented surge in severe vulnerabilities and AI-driven attacks, security expert Ami Luthvak and the hosts discuss why traditional static defenses and slow patching cycles are no longer sufficient. They examine how open-source models are catching up rapidly, why startups and enterprises must adopt continuous automated testing, and how defenders are beginning to fight back using automated red and blue agent pipelines.

  • Startups and companies managing AI keys must implement spending limits and robust security measures to protect against rapid LLM jacking attacks. 30:48
  • Security teams should shift focus from static code scanning to continuously testing unauthenticated external exposures with automated red agents. 27:35
  • New software companies should design their infrastructure from inception to support automated development, patching, testing, and DevOps pipelines. 9:56
  • Defenders must use multi-model pipelines that combine deterministic checks with progressively more expensive AI models to manage investigation costs. 15:16

Chapters

  1. 0:00
    Introduction and AI Attack Landscape

    Hosts introduce the episode and discuss how AI is accelerating zero-day attacks and changing the security game.

  2. 0:53
    Automating Blue and Red Teams

    Discussion on building automated pipelines for both defensive (blue) and offensive (red) security operations using AI agents.

  3. 2:11
    Red Agent: Continuous Offensive Testing

    Exploration of Wiz’s Red Agent, which simulates attackers to continuously probe and harden customer environments.

  4. 3:37
    Founders’ Playbook for AI-Enabled Security

    Guidance for startup founders on integrating security automation, patching, and AI-driven testing from day one.

  5. 43:01
    Scan for Good: Protecting Public Infrastructure

    Introduction to Wiz’s Scan for Good initiative aimed at helping public services and nonprofits identify and mitigate security risks.

  6. 45:13
    Closing Reflections and Future Outlook

    Hosts wrap up the episode, reflecting on the evolving threat landscape and the need for continuous, AI-powered defense.

Highlights

  • 0:41

    Most Security Work Isn't Securityinsight

    Ami argues that most traditional security team activities, such as compliance reporting and configuration checks, are not actual security work and should be stopped.

  • 2:11

    Black October and AI Exploitsdata

    Ami Luthvak describes October as a Black Swan event with an unprecedented volume and severity of vulnerabilities, highlighted by the PHP vulnerability found by AI.

  • 5:07

    Exploitation in Twenty Minutesinsight

    Open-source models are now capable of finding API mistakes and utilizing zero-day vulnerabilities to execute complete attacks in roughly twenty minutes instead of months.

  • 11:41

    Bitcoin Mining Vulnerability Analogyinsight

    Ami compares vulnerability discovery to Bitcoin mining, noting that as new models emerge, they continuously uncover fresh batches of critical bugs.

  • 15:42

    Blue Agent Investigation Pipelinestrategy

    Wiz uses a tiered investigation pipeline starting with deterministic checks, moving to cheap models, and ending with advanced models only when necessary to control costs.

  • 23:22

    Continuous Red Team Testingstrategy

    Because attackers gain new capabilities weekly, systems must be continuously attacked from the outside using the latest models before adversaries find vulnerabilities.

  • 41:43

    AI Surpassing Human Attackersinsight

    Unlike previous eras, cybersecurity is uniquely a domain where AI models are now faster and more capable than the best human attackers.

Quotes

  • “AI attackers that are actually going after companies, startups in scale and speed that we've never seen.”
    Unnamed speaker · 0:08
  • “The only companies that will be able to survive are companies that are...”
    Unnamed speaker · 1:01
  • “We should stop doing most of what we do and focus on finding critical risks, figuring out how we mitigate it, or at least put detections in, and also response readiness for it.”
    Unnamed speaker · 36:33
  • “The best defense is attack.”
    Unnamed speaker · 26:48

About this episode

AI isn’t just helping defenders find vulnerabilities. It’s giving attackers the speed and expertise to exploit them faster than many companies can respond. Wiz co-founder and CTO Ami Luttwak joins a16z General Partners Zane Lackey and Sarah Wang to unpack what he calls “Black October,” a wave of vulnerabilities and agentic attacks that he believes marks a turning point for cybersecurity. They break down PHP2Shell, why AI is collapsing the time between vulnerability discovery and exploitation, and what happens when open-source models put increasingly sophisticated offensive capabilities in more hands. Ami explains why startups are becoming attractive targets, particularly when they hold valuable AI credentials, and why yesterday’s secure architecture may no longer be secure as models improve. The conversation then turns to defense. Ami argues that companies need to continuously attack their own systems, automate investigation and patching wherever possible, design infrastructure assuming sophisticated attackers will get in, and rethink security stacks built for a slower era. His warning is simple: if attackers can move in minutes, defenders can no longer operate on human timelines.

Transcript

0:00
46:15

8,297 words · 4 speakers

Speaker names marked are inferred from speaking time and the show’s known hosts — they are not voice-verified.

Topics

  • Wiz
  • MITOS
  • Black October
  • A16Z

Page updated · transcript and summary are machine-generated